IBM faucets watsonx generative AI to assist modernize COBOL on mainframes

COBOL shouldn’t be a language usually talked about as a number one programming growth one, at the very least not in 2023. That wasn’t at all times the case. COBOL, which debuted in 1959, was a number one language within the earliest period of computing and there are nonetheless billions of strains of COBOL code operating manufacturing functions at present.

As we speak, IBM introduced a brand new initiative that makes use of the facility of generative AI giant language fashions (LLMs) to assist convey COBOL functions into the fashionable period. Among the many place that COBOL code continues to run is on IBM System Z (generally simply referred to a ‘Z’) mainframes. 

The brand new watsonx code assistant for Z service makes use of IBM’s watsonx LLMs for code growth to assist migrate COBOL utility to extra fashionable Java utility code. By modernizing functions incrementally on the mainframe utilizing gen AI, IBM goals to assist shoppers sort out expertise gaps and make the most of Java expertise whereas lowering threat.

IBM first detailed its watsonx product platform in Could throughout its Assume convention as an effort to construct out a collection of basis fashions for AI, designed for enterprise use instances.

“We’re actually seeing the usage of generative AI for code help changing into a key use case and rising market traits,” Skyla Loomis, VP for IBM Z Software program, stated throughout a press briefing. “Generative AI can assist builders to extra shortly assess, replace and check the correct code.”

Why updating COBOL is essential 

A key problem with COBOL code is that there’s a shrinking base of builders that really know keep it.

Loomis famous that roughly 84% of IBM’s Z mainframe shoppers are operating COBOL functions. As such, there’s a actual crucial to assist organizations modernize the code in a method that may be extra simply maintained. With the dearth of COBOL expertise, IBM took particular intention at coaching its AI so it might truly perceive the traditional program language.

Within the press briefing, Kyle Charlet, CTO for IBM Z Software program, defined how watsonx code assistant was educated to concentrate on COBOL code. Charlet stated that watsonx code assistant was initially educated on CodeNet, one of many largest code information shops on the planet. Along with the CodeNet code, he famous that IBM can also be actively coaching and tuning the watsonx mannequin.

“Enterprise Z COBOL is the place we’re additional tuning that mannequin and handing it a bunch of COBOL Java pairs in order that it understands precisely tune that mannequin,” he stated.

How generative AI transforms historical COBOL code into fashionable Java

The watsonx code assistant can be utilized to investigate, refactor, remodel and validate COBOL functions utilizing gen AI.

Charlet stated that the brand new providing can be utilized to “surgically extract” a logical enterprise service from a big monolithic COBOL utility. The watsxonx code assistant can then be used to generate a Java class hierarchy and remodel the extracted COBOL code to Java. To validate the transformation, the answer generates automated assessments to make sure the brand new Java code is semantically equal to the unique COBOL.

Charlet defined that the watsonx code assistant shouldn’t be doing a line-by-line COBOL syntax translation to Java. That might result in COBOL syntax expressed in Java, he famous, which in his expertise is basically unreadable and unmaintainable. The IBM strategy is to take the intention of the COBOL code and map it into Java code that is sensible.

“This Java needs to be recognizable and maintainable by Java professionals and fairly frankly, it’s,” stated Charlet.

Why code doesn’t lie or hallucinate, like textual content

A typical threat with gen AI know-how is that of hallucination, with outcomes that aren’t correct.

Whereas hallucination tends to be frequent drawback with AI textual content technology, Charlet argued that it’s much less probably with code normally for quite a few causes. He famous that with the interpretation of human language and potential hallucinations, a person won’t acknowledge the hallucination as an error. Textual content is topic to interpretation by people, code nevertheless works considerably otherwise than textual content.

Within the watsonx code assistant case, he defined that IBM validates the code that’s generated. The validation will instantly spotlight any hallucinations as a result of the code won’t run as anticipated if in any respect. As such, Charlet famous that whereas hallucination can doubtlessly happen with code technology, these hallucinations should not a matter of opinion and will be recognized and corrected.

“Code doesn’t lie,” stated Charlet. ” Code is one thing that you just observe and it’s a bunch of machine directions .”